Clinical risk prediction using language models: benefits and considerations

OBJECTIVE

The use of electronic health records (EHRs) for clinical risk prediction is on the rise. However, in many practical settings, the limited availability of task-specific EHR data can restrict the application of standard machine learning pipelines. In this study, we investigate the potential of leveraging language models (LMs) as a means to incorporate supplementary domain knowledge for improving the performance of various EHR-based risk prediction tasks.

METHODS

We propose two novel LM-based methods, namely "LLaMA2-EHR" and "Sent-e-Med." Our focus is on utilizing the textual descriptions within structured EHRs to make risk predictions about future diagnoses. We conduct a comprehensive comparison with previous approaches across various data types and sizes.

RESULTS

Experiments across 6 different methods and 3 separate risk prediction tasks reveal that employing LMs to represent structured EHRs, such as diagnostic histories, results in significant performance improvements when evaluated using standard metrics such as area under the receiver operating characteristic (ROC) curve and precision-recall (PR) curve. Additionally, they offer benefits such as few-shot learning, the ability to handle previously unseen medical concepts, and adaptability to various medical vocabularies. However, it is noteworthy that outcomes may exhibit sensitivity to a specific prompt.

CONCLUSION

LMs encompass extensive embedded knowledge, making them valuable for the analysis of EHRs in the context of risk prediction. Nevertheless, it is important to exercise caution in their application, as ongoing safety concerns related to LMs persist and require continuous consideration.

The particle size dependence of the oxygen reduction reaction for carbon-supported platinum and palladium

Model carbon supported Pt and Pd electrocatalysts have been prepared using a high-throughput physical vapor deposition method. For Pt, metal particle sizes are controlled between 1.5-5.5 nm over 100 electrodes of an electrochemical screening chip, allowing the oxygen reduction reaction (ORR) activity of the catalysts to be determined simultaneously. The ORR-specific current density is observed to increase with increasing particle diameter up to approximately 4 nm, at which point the activity begins to level off. The reduction in ORR activity for particles below 4 nm is accompanied by a concomitant increase in the overpotential for surface reduction. The resulting mass activity exhibits a maximum for particles with diameters of approximately 3.5 nm. These results are consistent with results published recently for high area carbon-supported Pt catalysts. For Pd particles, both the specific current density and the mass-specific activity for the ORR are observed to increase with increasing particle diameter, with no distinct optimum observed. The implications for the optimization of Pt- or Pd-based ORR catalysts for proton exchange membrane fuel cell (PEMFC) applications are discussed.

Reduction mammaplasty with the inferior pedicle technique: early and late complications in 371 patients

A review of reduction mammaplasties with the inferior pedicle technique in 371 patients over 10 years is presented and the complications associated with this particular technique are discussed. The mean age was 33.1 years (range 15-67), the average preoperative sternal notch to nipple distance was 30.4 cm (range 21-43), the mean breast tissue resected was 870 g per breast (range 250-1960) and the mean operating time was 3.1 hours (range 2.5-4). The overall complication rate was 11.4%. The incidence of specific complications in our series was: haematoma 0.3%, nipple and/or pedicle necrosis 0.8%, wound dehiscence 4.6%, fat necrosis 0.8%, carcinoma 0.5%, loss of sensitivity of the nipple 1.3%, hypertrophic scars 3.3%, dermoid cysts 0.3% and marked lower fullness 0.3%. 72% of the patients that became pregnant were able to lactate.